The modern data center is very demanding of its network infrastructure. Next-generation networks need to be more than fast. They also need to be smarter and provide deep insight. Speed seems like an obvious requirement, but there is more to network design than meeting the latest Gpbs in bandwidth. Next generation networks need to deliver the bandwidth improvements in such a way that they provide full-line rate, while ensuring that busy workload doesn’t starve other applications for bandwidth.
Support of Latest Standards
The first speed requirement of modern switches is that they support the bandwidth capabilities of next-generation networks. Today this means 32Gbps for fiber-channel (FC) connections and 40Gbps for Fibre Channel over Ethernet (FCoE) connections. Modern switches also need to support next-generation protocols like NVMe over Fabrics (NVMe-F). However, the reality is that upgrades to networks occur more slowly than the rest of the data center, so next-generation switches need to ensure that new protocols can co-exist with legacy protocols.
The second speed requirement of modern switches is for their architectures to be non-blocking. Each port on a switch may support 32Gbps or 40Gbps, but the internals of the switch may not support that bandwidth simultaneously across all ports. In an active environment where many of the switch ports are running full capacity, the switch may get overwhelmed, forcing traffic queues and causing some workloads to wait until traffic clears.
The non-blocking requirement is more critical in storage architectures that leverage all-flash storage. It is entirely possible to consume 100% of port bandwidth on the way in from the application server, while on the way out the flash-array can receive 100% of the port bandwidth.
Modern switches need to be rated so that all ports can be at 100% load. Non-blocking design requires more advanced components and more efficient switch software. Tier two white-box switches often lack the higher-end components required to deliver a non-blocking state. The organization ends up unable to populate existing switches fully, resulting in 50% of the ports in a data center going unused.
Part of a non-blocking architecture is to make sure that the ports themselves can deliver full line rate performance. The switch software and its associated hardware must outperform the port itself so that any overhead associated with the operation of the switch and the switch fabric doesn’t impact port performance.
There are more choices than ever available from switch vendors. IT planners need to be careful as they design next-generation storage network architectures. The temptation is to look at the port speed and assume that everything else is equal. IT planners need to look for switches that provide greater intelligence and telemetry data. They also need switches that automatically add themselves to the existing fabric and its associated zones. Finally, they need to make sure that the rated port speed is available to all ports, even if all ports are active.